<<

THE GERMANIC THIRD WEAK CLASS

JAY H. JASANOFF Harvard University

Germanic of the 3rd weak class form presents characterized by an alternation between predesinential *ai (e.g. . 3sg. habai1» and *a (e.g. Ipl. habam). These verbs are usually compared with the 'e-verbs' of Italic and Balto­ Slavic, but no IE present built on the stative suffix *-e- will account phonologi­ cally for the form of the suffix in Germanic. Instead, it can be shown that the characteristic Germanic paradigm results from the 'activization' of an older middle paradigm in which a 3sg. in *-ai « IE *-oi; cf. Skt. duh~ 'milks') was further suffixed by the productive active ending *-Pi « IE *-ti). 1. The inflection of the third class of weak verbs (exemplified by the 'to have': Gothic haban, hafa,! hab~, hebbian, habban) presents one of the classic problems in the historical morphology of Germanic. Not only do the verbs of this class show peculiarities in all the older , but they differ remarkably in their conjugation from one language to another, so that it is not at all obvious how the Common Germanic paradigm should be reconstructed. Given this diversity of forms, we will do well to begin with a short review of the morphological facts themselves. The situation is simplest in Old High German. The entire conjugation of hab~n is athematic (to the extent that this term still has any meaning), and is based on the single stem hab~-: 1sg. hab~, 3sg. hab~t, 3pl. hab~t, impv. 2sg. habe, opt. 3sg. habe, hab2e, prete 1sg. hab~ta, pres. pte. hab~ti, past pte. gihab2t. Except for the e of the stem, this paradigm is completely parallel to that of a 2nd class weak verb like 8alb~ 'anoint' (lsg. salbOm, 3sg. salb,jt, like habhn, hab~t etc.) The resemblance between the two classes extends to at least one set of forms which cannot be old. These are the longer optatives 3sg. hab2e, salb,je, 2sg. hab22s, salb62s etc., which seem to have been created analogically. The genesis of this formation, which will not concern us further in this paper, is discussed by Cowgill (1959: 11-12). The OHG pattern is not found in any other Germanic language. In Gothic the corresponding verb haban exhibits a systematic opposition between stem forms habai- (phonetically probably [habE· -]) and haba-. To the habai- group belong 2sg. habais, 3sg. habaijJ, 2pl. habaijJ, the corresponding imperatives (2sg. habai etc.), the weak preterit 1-3sg. habaida etc., and the past pte. habaijJs. The rest of the paradigm is indistinguishable from that of a normal strong verb: 1sg. haba, Ipl. habam, 3pl. haband (with impv. 3pl. habandau), opt. lsg. habau, 3sg. habai, passive 1-3sg. habada etc., pres. pte. habands. The Gothic alternation of stems hahai- and haba- is not completely isolated. It recurs almost exactly in Old Norse, where a typical 3rd class weak verb, vaka 'wake', is conjugated in the pres.

1 Here and below, 'Old Norse' will be understood to mean'Old Icelandic' unless other­ wise specified. This paper has benefited greatlyfrom extended discussions with Calvert Watkins, Warren Cowgill, Patrick Hollifield, and F. Roger Higgins. Needless to say, responsibility for errors is entirely my own. 850 THE GERMANIC THIRD WEAK CLASS 851 indic. as lsg. vaki, 2-3sg. vakir, Ipl. vQkum, 2pI. vaki'O, 3pl. vaka. Except for lag. vaki, each term is directly equatable with its Gothic counterpart; and vaki itself is almost certainly an analogical form, created to conform to the otherwise regular pattern lsg. = 3sg. minus -r (cf. 3sg. grefr 'digs', 1sg. gref; 3sg. heyrir 'hears', lsg. heyri). In the other moods and tenses the picture is the same, Old Norse in general showing forms in -i- < older *-e- < *-ai- where Gothic has -ai- (e.g. impv. sg. vaki, syncopated prete vakoa), and forms in -a- where Gothic has ·a- (e.g. info vaka, pres. ptc. vakandi). This inflection is documented in North Germanic from the 7th century; cf. the archaic runic compound wita(n)daha­ laiban 'protector of the bread', semantically parallel to OE hlajord 'bread-guard, lord'. (For ON hafa, segja 'say', and jJegja 'be silent', see §3 below). The greatest complexity is found in Old Saxon and Old English. In both lan­ guages, the 3rd weak class is a mere vestige of a category, the great majority of originally 3rd class verbs having been absorbed into the productive 2nd, or o-class. Old Saxon has carried this development to an extreme, and only two verbs, hebbian 'have' and seggian 'say', are usually considered to represent the unmixed type.2 Their inflection is characterized by an alternation between a stem in -e­ (or -a-), from older *-e- < *-ai-, and a stem in -ja-. The forms in -e- correspond in distribution to the -ai- forms of Gothic (e.g. 2sg. hobes, habas, 3sg. habed, habad) with the significant exception of the short preterit l-3sg. habda and the matching ptc. gihabd, which have the appearance of archaisms. The remaining forms, cor­ responding to those with the simple in Gothic, show general agreement with the 1st weak class: 1sg. hebbiu, pI. hebbiad, opt. 1sg. hebbie, pres. pte. *hebbiandi (cf. libbiandi 'living'), info hebbian; to these must be added the umlauted forms hahbiu, hahbiad etc., which occur with considerable frequency in the Monacensis manuscript of the Heliand. The situation is very similar in Old English. Habban, secgan 'say', and one or two others follow a paradigm essen­ tially identical to that of Old Saxon; the relevant West Saxon forms of habban (the absence of umlaut is no doubt secondary) are lsg. luebbe, 2sg. hcEjst, 3sg. "heefp, pI. habbap, opt. 1sg. hEebbe, pres. ptc. luebbende, info habban, prete l-3sg. ha3jde, past ptc. (3e-)ha3jd. In a number of other verbs, however, Old English has introduced a regularized paradigm (type Anglian lif3an 'live', 1sg. lijJe) which has no counterpart in Old Saxon. This type appears to have been created analogically on thebasis of the inherited 2sg. and 3sg. forms with *e stem vocalism (earlier *ai). It is itself subject to replacement by the productive 2nd class (cl. Cowgill 1959: 13-14). Although it displays an extraordinary degree of paradigmatic diversity in the older Germanic languages, the 3rd class is not a late category. Its archaic charac­ ter can be seen from the fact that a significant number of 3rd class verbs corre­ spond outside Germanic to formations built on a late IE deverbative suffix *-e-: typical examples are OHG dagen 'be silent', Lat. tacere; Go. witan 'watch', Lat. videre 'see', OCS videti, Lith. pavydeti 'envy' (both < *'ljide-, with the Balto­ Slavic lengthened zero-grade discussed by Kurylowicz 1956:291-3); OHG dolht 'suffer', Lith. tyleti 'be silent'; OHG haben, Lat. habere (from parallel roots

J OS libbian 'live' already represents a significant departure from the pattern of hebbian and seggian. I t is discussed in §4, below. 852 LANGUAGE, VOLUME 49, NUMBER 4 (1973)

*kap- and *ghabh-). This suffix, in which a 'stative' value is usually clearly recognizable, has enjoyed considerable productivity in a central group of IE languages comprising Germanic, Italic, Balto-Slavic, and Greek. It is very prob­ ably related to a homonymous suffix *-e-, equally well-established in IE but somewhat less prominent in Germanic, which serves to derive stative and in­ choative verbs from and . Denominative 3rd class verbs are synchronically important only in OHG, but their IE pedigree is assured by the important word equation OHG rothl, 'turn red', Lith. rudeti 'rust', Russian Church Slavic r1Jdeti s~ 'blush', OIr ·ruidi 'blush', Lat. rubere 'be red', all pre­ supposing a verbal stem *rudh-e- < adj. *rudh-ro- (cf. Watkins 1972). It is hardly surprising that scholars have been virtually unanimous in connecting both Germanic types with the 'e-verbs' of other IE traditions. But while the fact of the relationship has been universally admitted, there has been very little agreement on how the morphological details of the Germanic situation are to be explained. Accordingly, it will not be amiss to survey a few of the IE and Pre-Germanic formations most commonly assumed to have played a role in the prehistory of the 3rd class. ATHEMATIC PRESENTS IN *-e-. The OHG inflection hab~, -~s, -~t etc. suggests the possibility of an inherited athematic present in *-e-mi, *-e-si, *-e-ti. Such a formation is actually found in Hittite verbs of the denominative type (e.g. dannattet 'became empty', dannattas 'empty'; cf. Watkins 1972) and may lie at the root of a Lat. 3pl. like lucent 'shine' < *-enti if this is not simply analogical. But while perfectly acceptable from an IE standpoint, the hypothesis of an athematic e-conjugation is not easy to reconcile with the Germanic forms. The view that OHG hab~t continues a Common Germanic *habejJ(i)3 has most re­ cently been argued, and with considerable resQurcefulness, by Polome (1967:83­ 92), who is quite correct in noting that there are no insurmountable obstacles to his theory as far as West Germanic is concerned. On the other hand, his iden­ tification of *habejJ(i) as the source of Go. habaijJ seems wrong, and leaves the simple thematic forms haba, habam, haband unexplained.4 Perhaps the most telling objection to his view is the state of affairs in Old Saxon: Why, if Old Saxon in­ herited 1sg. *haben < *-em, did it replace this form by the wholly irregular hebbiu while retaining-jf not actually creating-the parallel 1sg. salbon in the second class? It seems more likely that 1sg. *haben never existed in Old Saxon; and that the athematic (and thoroughly 'regular') appearance of OHG habbn,

3 Here and below, I have somewhat arbitrarily chosen to ignore the effects of Verner's Law on the Common Germanic verbal endings of the 2nd and 3rd persons, singular and plural. In the reconstructions that follow, these endings will appear with a uniformly ­ less spirant; the actual situation, of course, must have been considerably more complicated. 4 Even assuming that haband is the regular reflex of CGmc. *habenjJ(i)-and this is by no means certain-I find it hard to understand why this 3pl. form should have served as the starting point for a new and anomalous paradigm, when in the parallel 2nd class the regular 3pl. *salband < *-onjJ(i) has simply been remade to salbond. There seems to be little to rec­ ommend Polome's suggestion (p. 90) that 1sg. haba derives phonologically from *habem (i). In fact, the inscriptional forms Lubaini (= Go. lubains 'hope') and Vanaenia (= Go. wanains 'lack') argue rather forcefully that the ai of Go. habaijJ etc. is an etymo­ logical diphthong (for discussion see Krahe 1961:37-9). THE GERMANIC THIRD WEAK CLASS 853

-es, -et is an innovation based on the inherited forms where e, of whatever origin, is old. But if this is so, then the assumption of an athematic conjugation in *-e- is superfluous: there are no forms in any Germanic language which require it. THEMATIC PRESENTS IN *-ei6/o-. The equation 3sg. Go. habaip = OHG habet = OS habed (-ad) points most naturally to a Common Germanic 3sg. *habaijJ(i) , from which the development of *habejJ in North and West Germanic would be entirely regular. Many scholars, before and after Brugmann, have in turn sought to derive *habaijJ(i) from *kapei.eti, an e-present extended by the productive thematic suffix *-ie/o-. While not necessarily an IE formation, presents of this type are common in Balto-Slavic, where they are found only with denominatives, e.g. Lith. rudeti, rudeja; Russian Church Slavic rodeti S(J, r'bdejet'b S(J (but Lith. srnirdeti, srnirdi 'stink', OCS srnr'bdeti, srnr'bdit'b). Even though'no other examples are known for the putative development *-eie- > *-eji- > *-ei- > *-ai- in Ger­ manic, there are no very serious phonological obstacles to explaining the ai-forms in this way. The real difficulty lies in the cases where the thematic vowel is IE *0, i.e. in the 1sg., 1pl, and 3pl. Neither Go. haba, -am, -and nor as hebbiu, -ad can represent the normal phonological continuation of a pre-Germanic *habeJo, *habejame(s), *habejanpi; thus, while it is not impossible that a suffix *-eitj/o­ may have played some role in the creation of the 3rd weak class, this suffix alone will not account for the attested paradigm in any Germanic language. One is inevitably led to speculate on the possibility of admixture \vith one or more other inflectional types, but explanations of this kind have proved difficult to motivate. PRESENTS IN *e(i)-/-i-. Certain scholars, most recently and notably Schmid (1963:94-7), have taken as original the alternation of 1sg. as hebbiu, OE hcebbe with 3sg. as habed, OE heefjJ, comparing these forms with Lith. 1sg. smirdiiu, inf. smirdeti; Wagner (1950:50-54) would add Hitt. 3sg. lJalzai, 3pl. lJ,alziyanzi 'call'. Schmid's reconstruction assumes an ablauting suffix *-ei- with zero-grade *-1:-, both optionally followed by the thematic vowel and so distributed as to yield a paradigm 1sg. *kapio, 2sg. *kapeiesi, 3sg. *kapej,eti, 1pl. *kapj,ome, 2pl. *kapej,ete, 3pI. *kapj,onti. In IE terms this is an unlikely inflection at best, and the extra-Germanic evidence adduced in its support will not stand scrutiny. As Kurylowicz (1964:79-84) has now shown, the BaIto-Slavic 'semi-thematic' i-presents ultimately reflect a transformation of the IE ,6 while the Hittite forms are in all probability an Anatolian innovation (cf. Jasanoff 1968: 19-55).

6 I am indebted to Warren Cowgill for pointing out several difficulties with Kurylowicz' explanation as presented in the work cited. In particular, there seem to be serious phono­ logical objections to deriving the theme vowel of the Baltic forms from an i-diphthong shortened in final position; Latvian and Old Prussian evidence points rather to Common Baltic short i. In my opinion, this i has been analogically generalized from an athematic 3pl. ending *-int < *-V-t or *-ir < *-r, but a detailed presentation of the arguments which support this conclusion must be reserved for a separate study. For present purposes, it is enough to say that Kurylowicz' theory, while needing some revision in detail, is substan­ tially correct in explaining the Balto-Slavic i-verbs as the reflex of an athematic formation built directly on the root. Whether this formation should be thought of as identical to the IE perfect (rather than, e.g., the athematic middle) is an important and still unresolved question; it does not, however, affect the interpretation of the Germanic forms in any im­ mediate way. 854 LANGUAGE, VOLUME 49, NUMBER 4 (1973)

Even on a-priori grounds, there are good reasons for regarding the Old Saxon and Old English forms in *-ja- as secondary. Old Saxon and Old English are precisely the languages in which the 3rd class, as a moribund category, is most open to influences from the other weak types; note such borrowings from the 2nd class as OE 2sg. haJast, 3sg. haJap, OS 2sg. libos, 3sg. libod (cf. libbian 'live'), or forms like OE info liJJan < *libaijan, pI. liJJaj> < *libaiJanp etc., ultimately created on the model of the 1st (nerian) class. The genesis of the characteristic Northern West Germanic paradigm lsg. *habjo, 2sg. *habais, 3sg. *habaip, pI. *habjanp is a problem that will concern us at greater length in §3; for the moment it is enough to observe that neither internal evidence from Germanic nor com­ parative data from any other IE language suggests that this type is anything but the late creation of one specific dialect area (cf. the conclusion reached on other grounds by Polome, 85). It emerges, then, that all three major have innovated significantly in their treatment of the 3rd class. We can now assert that the Com­ mon Germanic paradigm is essentially preserved in Gothic and Old Norsc, and is to be reconstructed in the present as lsg. *habo, 2sg. *habais(i), 3sg. *habaij>(i), 1pl. *habame(s), 2pI. *habaij>(e), 3pI. *habanj>(i). The mere agreement of Gothic and Old Norse, as against West Germanic, in showing this inflection, is strong prima-facie evidence for its Common Germanic character. But as a matter of fact-and the importance of this seems not to have been properly appreciated before-West Germanic itself directly attests reflexes of at least one diag­ nostic member of the Gothic-Norse paradigm, the lsg. *habo. The regular lsg. of habban in the Northumbrian dialect of Old English is haJo (-u), a form which notably contrasts with West Saxon hcebbe. HaJo is matched by sre30 (= WS s6cge); the usual pattern in the Northumbrian Lindisfarne Gospels ia lag. haJo (sce30), 2sg. hceJes (sre3es), 3sg. hceJep (83J)ep), pI. habbap (smcgap).6 These forms have been explained as due to analogy with the normal strong paradigm (e.g. by Flasdieck, 23), but this is most unlikely in view of the frequency with which Northumbrian maintains the geminatedjungeminated alternation in regular ja-presents. Further evidence that haJo and sreJo are archaisms comes from Old High German, where Tatian (scribe 'Y) preserves the completely iso­ lated lsg. forms habu (2X), sagu (IX) beside normal haben, sagen < habbn, sagbn. These are not, as has sometimes been asserted, back formations froms the problematic 2sg. hebis, segis, 3sg. hebit, segit, which do not occur in Tatian. Habu and sagu are found embedded in the normal ~-paradigm in two interesting contexts: 87.5 Antuurtanti daz uib inti quad: ni habu gomman. Tho quad iru der heilallt: uuola quadi thaz thu ni habes gomman. 'Respondit mulier et dixit: non haheo virum. Dicit ei Jhesus: hene dixisti, quia non habeo virum.' 88.10 Oha ih giuuizscaf sagu (corr. f: sagen) uon mir min giuuizscaf nist uuar. Ander ist der dar giuuizscaf saget uon mir ... 'Si ergo testimonium perhibeo de me, testimonium meum non est verum. Alius est qui testimonium perhihet de me ... '

6 Cf. further lsg. liofo (Lindisfarne), lifo (Rushworth II). Here and elsewhere, Old Eng­ lish forms are quoted after Flasdieck 1935. THE GERMANIC THIRD WEAK CLASS 855

The pattern habu, habes, habet (sagu, sages, saget) which emerges from these passages accords perfectly with Northumbrian hafo, hcefes, hcefejJ and Go. haba, habais, habaijJ. It is hard to escape the conclusion that this inflection is 01d. 7 The distribution of forms in -a- and -ai- in the Gothic (and Common Ger­ manic) 3rd class paradigm is not easy to motivate historically. A recent attempt to go beyond vacuous 'mixture' theories is that of Bennett (1962:136-41), who proposes to derive all the principal Gothic forms from a single earlier type in *_~e/o-: 1sg. *kapaio, 2sg. *kapajesi, 3sg. *kapaieti > CGmc. *habo, *habais(i), *habaijJ(i) > Go. haba, habais, habaijJ. Bennett considers *a the zero grade of the suffix *-e- (i.e. *-e~l- ); his rules call for the loss of antevocalic *ai except before *i < *e, where *ai > a. This is not a convincing reconstruction. The suffix *-e­ is notable for not showing a clear zero grade in any other IE language. Besides being completely ad hoc, Bennett's rules aSsume the retention of *a as a separate vowel until well into the Germanic period-an assumption which other evidence, such as the form of the word 'daughter' (Go. dauhtar etc. < IE *dhug(h)ater-) , speaks rather strongly against. For us the main interest of Bennett's theory lies not in the details of his reconstruction, but rather in his explicit recognition that the Germanic paradigm *habo, *habais(i) etc. CANNOT be explained on the basis of an inherited present in *-e- or *-ei(e/o)-. There should be nothing too surprising in this result, for there is very little evidence that the deverbative suffix *-e- played a role in the formation of an IE present. In Greek the main older function of the suffix is to provide , de­ noting entry into a state, to otherwise characterized presents-ekhdren to kha1,ro 'rejoice', ekden to ka1,o 'burn', emdnen (post-Hom.) to mainomai 'rage'. This is the usage from which the classical passive in -the- develops (cf. Chantraine 1925:105, and Prevot 1935:86-100). Particularly in view of its expanding role in the formation of the future (skhiso < ekho 'have', pitMso < peitho 'persuade') and perfect (nenemeka < nemo 'distribute', memeneka < meno 'wait'), the exclusion of -e- from the present system is remarkable. This aspect of the Greek situation recalls the state of affairs in Baltic and Slavic, where the suffix *-e- supplies preterits (aorists) and to 'stative' verbs whose pres­ ents typically continue a formation akin to the IE perfect. Parallel to Gk. aor. emanen, pres. mainomai, Lithuanian has inf. mineti 'remember', prete 3sg. minejo, but pres. 3sg. mini < pf. 3sg. *mone-i; the matching OCS forms are inf. mbneti, aor. 3sg. m1Jne, pres. 3sg. manit'b < *mone-i + 3sg. pronoun t'b (cf. Kury­ lowicz 1964:81-3, Watkins 1969:222-5). Only Latin, with e-presents like habeo 'have', maneD 'remain', taceo 'be silent' etc. departs significantly from the Greek and Balto-Slavic pattern. It is far from clear that this situation is old. The Latin 2nd conjugation is notoriously a mixed category, comprising in addition to the habeo type both denominal formations (e.g. albeo 'be white', sene0, be old',

6 Tubeo 4be red') and iterative- in *-ei / o- (e.g. moneo 'warn', noceo 'harm', doceo 'teach'). In all probability the e-inflection of deverbal statives in the is an innovation of Latin; a stative present like habeo follows the model of denominal rubeo (and/or moneo, where -e- < *-eie-) in

7 The later forms habo and sago (Notker etc.) do not seem to belong here. They can easily be motivated analogically, and are best omitted from the present discussion. 856 LANGUAGE, VOLUl\1E 49, NUMBER 4 (1973) precisely the same way that the dialectal Lith. pres. 3sg. rnineja (for mini < ·mone-i) follows that of denominal 3sg. rudeja 'rubet'. These facts bear importantly on the problem of the 3rd weak class. We have seen that while the stems *habe-, *wite-, *mune- « *kap-e-, *1Jid-e-, *mon-e-) may be old in Germanic, it is difficult to interpret them as directly underlying the corresponding 3sg. presents *habaijJ(i) , *tvitaijJ(i) , *munaijJ(i). The con­ clusion suggested is that in the oldest Germanic, as in the oldest Greek and Balto-Slavic, the stative suffix *-e- was restricted to one or more non-present functions. Such a function might well have been that of aorist, but the exact original value of the e-extended stem forms is of no special importance to the present discussion. My main assumption will simply be this: at some time in the pre-history of Germanic, a verb like 'have' was characterized by an opposition between a stem *habe-, serving in a non-present function, and an independently formed second stem, unsuffixed by *-e- and supplying the present. The origin of the historical -ai-/-a- present may therefore reasonably be sough~ in some semantically and formally appropriate IE category which, like the Greek in­ trans. pres. middle in *-i.e/o- (mainomai) and the Balto-Slavic perfect (mini, mbnit'b), has come to be secondarily associated with the extra-present formation in *-e- (cf. emdnen, minejo, m'bne). Such a category can indeed be found in Germanic, as I shall attempt to demonstrate in the following discussion. 2. The Gothic verb hahan 'hang' (tr.) agrees with its in the other old Gmc. languages (OHG kahan, OE hon etc.) in belonging to the 7th strong (reduplicating) class. In meaning, hahan contrasts with the partly homonymous 3rd class weak verb hahan 'hang' (intr.); the semantic relation between the two is the same as that which Greek expresses by an opposition of voice (krhnnemi, kremdnnumi vs. kremamai) and Latin by the presence vs. absence of the suffix *-e- (pendo vs. pendeo). The functional opposition between strong hahan and weak kahan, which recurs exactly in OHG hahan vs. hangen, gives the appear­ ance of an archaism; note that the later languages typically utilize more 'modem' devices to express the same contrast-e.g. earlier New High German, which opposes hiingen (tr., < iterative-causative *hangjan) to hangen (intr.) The Com­ mon Germanic 3rd class weak verb *hangai-/*hanga- 'hang' (intr.), 3sg. *hangaijJ(i), is not etymologically isolated. It is clearly to be connected with the thematic deponent sdnkate 'hesitates, doubts', a verb first directly at­ tested only in the Brahmanas, but presupposed by the deverbative Rigvedic hapax sanku- 'hanging'. Sd?ikate in turn is related to Lat. cunctor, -ari 'hesitate', to a lost deponent most naturally to be restored as *concor, -ii. Hittite gangalJlJi 'I hang' (tr.) is a further ; though transitive and active, this verb resembles other thematic members of the lJi-conjugation in seeming to presuppose the former existence of a thematic middle 1sg. *gangafJfJa 'I hang' (intr.) (cf. Meid 1971:77-8, Watkins 1969:117). The root *konk-thus appears to have been inflected in the middle voice from a very early period; outside Ger­ manic no fewer than three independent IE traditions attest directly or indirectly to a thematic middle paradigm with intransitive meaning. This fact can be ex­ ploited to provide an explanation for the formally problematic 3sg. *hangaijJ(i). I propose to see in 3sg. *hangaijJ(i) the reflex of an archaic 3sg. middle in THE GERMANIC THIRD WEAK CLASS 857

*-ai < IE *-oi (*-0 + deictic *i), suffixed secondarily by the normal Gmc. 3sg. pres. termination *-jJ(i). The position of the ending *-o(i) in the scheme of the IE verb has recently been discussed exhaustively by Watkins (1969, passim), who shows how an inherited 3sg. in *-0 (apophonically an alternant of the 3sg. perfect in *-e) is renewed as *-to, *-eto, and *-oto in the majority of IE traditions. These dental-extended forms of the basic ending, themselves subject to optional enlargement by the particles *r and *i, underlie the 'classical' 3sg. middles in *-to(i), *-eto(i), *-oto(i) (Indo-Iranian, Greek, Germanic) and *-tor(i), *-etor(i) , *-otor(i) (Hittite, Tocharian, Italic, Celtic). However, the archaic t-Iess forms *-o(i), *-or(i) are not purely hypothetical constructs; they are preserved intact in 3sg. middles, athematic and thematic, in Indo-Iranian (Vedic athematic duhe 'milks', later dugdhe, impf. dduha(t); thematic sobhe 'praises', later 80bhate) and Hittite (athematic esa, esari 'sits' beside prete estat; thematic neya, neyari 'leads' beside neyatta). Cf. further Umbrianferar 'feratur' and OIr. passive berar 'is carried', beside deponents like do·moinethar 'thinks'. Within Germanic, where the IE middle is continued in the Gothic passive (e.g. 1-3sg. bairada 'am, is carried' < *berajJai), a similar replacement must be assumed: the 3sg. desinence Go. -ada, CGmc. *-ajJai ultimately points to an older ending *-a (or suffixed, *-ai) which is nowhere attested as such. However, interesting indirect evidence for a 3sg. in *-ai is still to be seen in the peculiar and isolated use of the Gothic ending -ada in the 1sg., a feature which recurs in OE hatte 'am, is called' = Go. haitada. Since IE *a and *0 both appear as a in Germanic, the inherited ending of the 1sg. middle *-ai < *-oi must early have fallen together-at least in the athematic type-with the 3sg. *-ai < *-oi. Following their phonological merger, the two endings were treated as identical, and the subsequent replacement of *-ai by *-ajJai affected the 1st and 3rd persons equally (cf. Watkins 1969:138). It is therefore reasonable to assume dentalless *hangai, *berai etc. as the oldest forms of the 3sg. middle in Germanic. Such forms, I shall argue, have had a twofold development in the historical languages-yielding, on the one hand, the Gothic (and Common Germanic) passive in 3sg. -ada « *-ajJai); and on the other, via a process now to be described, the Common Germanic 3rd weak class in active 3sg. *-aijJ(i). At the earliest stage of Germanic accessible to us, the middle 3sg. *hangai must have been bifunctional: like the Greek middle, it invested the root meaning 'hang' with intransitive value, and at the same time optionally signaled the grammatical passive to the transitive active 3sg. *hanhijJ(i). This situation, of a type frequently encountered in older IE lan­ guages, was conducive to the introduction of a new formal contrast to differen­ tiate the two functions. In the passive (in Kurylowicz' terms, 'primary') func­ tion,8 the 3sg. hanga(i) was renewed via the replacement *-a(i) > *ajJa(i) as the ~ In this respect the Germanic hierarchy of functions is the exact opposite of that in Indo-Iranian, where the dentalless ending *-ai (Skt. -e) is typically specialized as a mark of the passive (skt. groe 'is praised', bruve 'is said', srove 'is heard'). There is no real contra­ diction here. In Indo-Iranian the passive function of the old middle is taken over by a new category, the derived passive in -yd-; the use of inherited middle forms with the same value is a marginal archaism. In Germanic, on the other hand, the (originally peripheral) passive signification of the IE middle is developed as its sole productive function, and it is as a mark of the passive that the inherited middle ending of the 3sg. undergoes formal renewal (cf. Watkins 1969:88 and Kurylowicz 1960:79). 858 LANGUAGE, VOLUME 49, NUMBER 4 (1973)

familiar Germanic passive of the handbooks, *hangajJai, Go. *hahada (with regular loss of ). The spread of the dental ending thus came to be associated with the growth of a new grammatical category, the pas­ sive, while the dentalless ending was restricted to intransitive ('secondary') func­ tions alone. With the subsequent loss of the middle as an autonomous category in Germanic, the status of the archaic *hangai, now re-interpreted as a derived present unmarked for voice, would have become precarious in the extreme. Like over-short 3rd singulars generally, it was liable to be resegmented as a bare stem form with desinence zero, Le. as hangai+0 instead of the historically correct *hang+ai. But this new segmentation made possible the accretion of a new and overt mark of the 3sg. in place of the null ending. The result was the suffixation of *hangai by the unmarked (active) 3sg. desinence *-jJ(i), and the creation of CGmc. 3sg. *hangaip(i)-the form directly accessible to us through compari­ son. Once established as a 3sg: active in form, *hangaip(i) was free to serve as the starting point for a new active paradigm. It is possible to give a sketch of how the transformation from a middle to active might have proceeded outside the 3sg.: Germanic, like Hittite, Tocharian, and the neighboring Baltic, appears to have inherited from dialectal IE a thematic conjugation with persistent o-color of the thematic vowel. This is directly visible in the Go. passive 1sg. haitada, 2sg. haitaza, 3sg. haitada, pI. haitanda; cf. further Hitt. middle 1sg. neyalJ,lJa(ri), 2sg. neyaua(ti), 3sg. neya(ri), 3pl. neyanta(ri); Toch. deponent 1sg. A miiskamar, B maskemar 'I am, ich befinde mich', 2sg. A maskatiir, B masketar, 3sg. A maska­ tar, B miisketiir, 3pl. A maskantar, B maskentar;9 Lith. 1sg. vedu < *-0 'lead', 2sg. vedi < *-ai, 3sg. veda etc. Following Watkins (1969:213), we may display the terminations of this type, so far as known, as follows: 1sg. *-0-~20 1pl. -- 2sg. *-0(-t~20) 2pl. -- 3sg. *-0 3pl. *-o-r (?) Replacing the unattested r-ending of the 3rd plural by its successor *-nto, and adding the particle *i of the hic et nunc, we obtain, after regular phonological changes, 1sg. *hango (-oi ?) 1pl. -- 2sg. *hangatO 2pl. -- 3sg. *hangai 3pl. *hanganjJai This is the paradigm which is 'activized', via the 3sg. *hangai, in the manner described above. Re-interpreted as a bare stem, *hangai engenders the new 3sg. *hangaip(i) and the corresponding 2sg. *hangais(i). These active forms in turn lead to the elimination of the remaining middle endings: the 1sg. and 3pl. appear

9 This verb is representative of the Krause-Thomas 3rd present class (1960:200-201). The correspondence a(A) = e(B) points to IE *e or *0 as the predesinential vowel in this type, but only *0 will explain the consistent non-palatalization of the root-final consonant. 10 I see no need to assume a 1st person ending *-oi for Germanic; on the putative 1sg. *haitOi'I am called', see below. In the 2nd person it is simplest to reconstruct *-ai, an end­ ing which recurs in Baltic; but this choice is of no particular importance. In principle we could just as easily operate with a form *hangasai, or even *hangapai. THE GERMANIC THIRD WEAK CLASS 859 with the regular active endings as *hango and *hanganjJi respectively, while the 1pI. and 2pl., discarding older middle forms, become *hangame(s) and *hangaijJ(e). Note that the 2pl. shows an assimilation in stem form to the 3sg., in conformity with the normal pattern elsewhere in Germanic (e.g. Go. 3sg. bairijJ = 2pl. bairijJ etc.) The result is a new paradigm, from which the histori­ cal forms of *hangana(n) 'hang' (intr.) are derived: lsg. *hango lpl. *hangame(s) 2sg. *hangais(i) 2pl. *hangaijJ(e) 3sg. *hangaijJ(i) 3pl. *hanganjJ(i) The importance of the verb 'hang' for the history of the 3rd class as a whole should now be quite clear. Comparative data indicate that the root *konk- in­ herited a thematic middle inflection from late PIE, while there is no evidence outside Germanic to support the assumption of an e-extended stem *konk-e-. Our derivation of *hangaijJ(i) and its associated paradigm from an old thematic middle is consistent with these findings, and at the same time provides a fairly natural explanation for some of the historically troublesome details of the 3rd weak class inflection. This suggests in turn that the hypothetical renewal *hangai ~ *hangaijJ(i) , rather than being an isolated analogical replacement affecting only one lexical item, may in fact be part of a quite general Germanic process which has led to the creation of the characteristic, and hitherto unex­ plained, present type in *-ai-/*-a-. It should be noted that no part of the development assumed for 'hang' lacks clear typological parallels elsewhere in IE. We have already seen examples in Indo-Iranian (s6bhe ~ s6bhate) and Hittite (neya ~ neyatta) of a formal renewal in these languages directly comparable to the Germanic replacement of *hangai by *hangajJai in the passive function. A third branch of the family, Celtic, shows evidence of the same replacement, and agrees further with Germanic in special­ izing the endings *-0 and *-to in different values-the first as the normal the­ matic passive 3sg. (OIr. berar 'is carried'), the second as the deponent 3sg. (OIr. do·moinethar 'thinks'). The Germanic replacement of middle by active forms outside the 3rd singular (3pl. *hanganjJai ~ *hanganjJ(i) etc.) is quite unremarkable in itself, and ultimately comparable to the loss of deponent inflec­ tion in Late Middle Indo-Aryan and Vulgar Latin (cf. already in Old Latin moro for moror 'delay', nasco for nascor 'be born', pacisco for paciscor 'bar­ gain', etc.) The special aspect of the development which may seem bizarre-the direct application of the ending 3sg. *-jJ(i) to the already complete middle 3sg. *hangai-has in fact an almost exact parallel in the history of Vedic like dduhat 'milked', dSayat 'lay': these, as Wackernagel showed long ago, owe their final -t to a secondary amplification of older dentalless forms dduha, *d8aya. There is a second and even more striking example of the same process in Hittite, where, particularly in the later language, certain apparently thematic lJ,i-eonju­ gation verbs like tarna- 'leave' (lsg. tarnalJlJi, 2sg. tarnatti, 3sg. tarnai) tend to develop an alternative mi-conjugation inflection. The starting point for the new paradigm is the dentalless 3sg. tarnai, which is directly suffixed by the produc­ tive 3sg. ending -zi < *-ti; the resulting mi-conjugation forms *tarnami (cf. isgami < isgalJ,lJi 'I anoint'), 2sg. tamasi, 3sg. tarnaizzi are quite reminiscent of 860 LANGUAGE, VOLUME 49, NUMBER 4 (1973) the reconstructed Gmc. *hango, *hangais(i), *hangaijJ(i). A morphological re­ placement of this kind appears to be responsible for the creation of a consider­ able number of Hittite verbs in 3sg. -aizzi (for discussion cf. Watkins 1969: 102, Jasanoff 1968:69-92).11 It would be well to observe, before proceeding further, that Germanic pre­ serves a well-known example of a thematic middle which has been refashioned to an active in a manner strikingly similar to that sketched above. The Old Norse strong verb heita 'bid, call' has two presents: one the regular lsg. heit, 2-3sg. heitr etc.; the other, in the special meaning 'be named, be called, heissen', lsg. heiti, 2-3sg. heitir etc. The latter forms are clearly medial in origin and con­ nected in some way with Go. 1-3sg. haitada, OE 1-3sg. hatte, both regularly reflecting a CGmc. passive *haitapai. Neither heiti nor heitir can directly con­ tinue *haitapai, however; the expected reflex of this form in Old Norse would be *heitti. Accordingly, many scholars have seen in lsg. heiti the persistence of an IE middle desinence *-ai « *-~2o-i), as in Skt. bhdre, or (more plausibly) *-oi « *-o-~2o-i)-as, formally at least, in Old Hittite tarnalJbe 'I leave' (lsg. lJi-conjugation) (for discussion, see Cowgill 1968:24-31). Such scholars implicitly regard 3sg. heitir as an analogical form, created by adding the productive ending -r to the inherited lsg. heiti. The difficulty with this view lies in the fact that, as already noted, elsewhere in Old Norse the 1sg. tends to be remade on the model of the 3sg., not vice versa. From both a typological and a purely Norse standpoint, an analogical progression heiti ~ heitir would be an anomaly, com­ pletely at variance with later re-formations like grefr 'digs' ~ gref and vakir 'wakes' ~ vaki. We shall do better to seek an explanation for heitir on its own terms; and such an explanation is in fact available if we see in heitir the continu­ ation of a pre-form *haitaijJ(i), remade from an old dentalless 3sg. *haitai via suffixation of the ending *-p(i). In this way we account directly for the fact that the present tense paradigm of heita is completely identical to that of a 3rd class weak verb. The twofold development of *haitai to *haitaip(i) in North Ger­ manic, *haitapai in East and West Germanic, would then reflect the real seman­ tic ambiguity inherent in the notion of 'being called', which is variously treated as a middle or as a true passive in the individual IE languages. 12

11 However, it must be noted that, outside the 3sg., the correspondence between Hittite and Germanic is not exact. Forms like 3sg. tarnaizzi were identified in Hittite with inherited 3rd singulars like 1)atraizzi 'he writes', where -aizzi apparently continues older *-ai,eti; thus *tarnami, tarnasi etc. are not simply thematic, but rather imitate lJ,atrami, 1;atraBi etc., where the connecting vowel a (usually written with the scriptio plena) reflects original *-ajp-. 11 The Runic lsg. forms haite and haiteka, found as early as the fifth century (haite, Kragehul, early 6th century; h(a)ite, Jursberg, 500-550; ha(i)teka, Lindholm, early 6th century; haitika, Seeland II, 450-550; cf. Krause-Jankuhn 1966) do not constitute a real argument against the assumption of an original3rd class lsg. *haito, even though the lsg. ending -u < *-0 is retained in strong verbs during this period. In view of the absence of an attested lsg. to a 3rd class weak verb in early Runic Norse, it is perfectly possible that the inherited lsg. in *-0 had already been remade to *-e after the 3sg. in *-ejJ < *-aijJ. The model for this development would have been supplied by the 2nd weak class (cf. Go. lsg. salbo, 3sg. salbojJ) , where the rule lsg. = 3sg. minus *-jJ(i) is Common Germanic; given the second­ ary creation of OHG lsg. habem after the model of lsg. salbtJm, a like analogical origin for THE GERMANIC THIRD WEAK CLASS 861

3. In conformity with the preliminary conclusion reached at the end of §l, 1 am no,v prepared to make the following claim: THE THIRD WEAK CLASS ULTI­ MATELY CONTINUES AN EARLY GERMANIC TYPE IN WHICH A THEMATIC MIDDLE INFLECTION IN THE PRESENT CONTRASTED WITH FORMS BUILT ON A STEM IN *-e­ IN ONE OR MORE NON-PRESENT CATEGORIES. If this is correct, it means that at some time in its pre-history, Germanic came secondarily to pair its inherited (late IE aorist?) stems in *-e- with presents of an etymologically entirely un­ related type.13 This, of course, parallels completely the situation ,ve have seen in Balto-Slavic, where the present of 'e-verbs' is supplied by a congener of the IE perfect (Lith. prete minejo, turejo; pres. mini, turi < *mone-i, *ture-i). But it should hardly be necessary to emphasize that I do not mean to imply that all, or indeed even most, attested 3rd class weak verbs were originally middle or deponent. Thus verbs meaning 'have' are not characteristically middle in the older IE languages; and there is little reason to believe that, in the oldest stages of Germanic, the ancestor of Go. haban, OHG haben etc. was an exception. By the same token OCS vidit'b 'sees', Lith. pavydi 'envies' cannot be old as perfects; the inherited perfect of the root *y,eid- is still preserved with its original mean­ ing in the archaic Slavic Isg. vede '1 know' < *1Joidai. Both cases must be ex­ plained from the fact that the association of stative stems in *-e- with middles in Germanic and perfects in Balto-Slavic has become productive. Starting from relatively small beginnings in both traditions, a new pattern has established itself and spread. In Balto-Slavic the -aorist stems *1J'ide-, *ture- auto­ matically entail perfect-presents *1J'idei, *turei (> -vydi, turi); in Germanic the corresponding stems imply middle (deponent) presents *witai, *habai (> Go. witaip, habaij;).14

lsg. haite would hardly be surprising. The tendency of the lsg. to be remade on the model of the 3sg. is old in any event; cf. lsg. gibu for gebu on the same bracteate as the form lsg. haitika. In the case of heita in particular, the creation of the new lsg. haite would have been favored by the need to distinguish the active and middle senses of the inherited *haito, and perhaps also by the ambiguity of the common form haiteka (either *haito + *eka or *haite + *eka). The interpretation of heita offered here should lead us to expect a 3rd class weak prete lsg. *haitaidom, standing in the same relation to the present *haitaip(i) as Go. habaida stands in relation to habaip. No such form can be traced in Gothic or Old Norse, where the preterit of this verb in the intransitive sense is simply supplied by the + past parti­ ciple. Yet *haitaidom, I would suggest, has survived; it is the direct antecedent of OE hatte, pI. hatton, the anomalous preterit to the 'passive' pres. I-3sg. hatte discussed above. Taken together in this way, Old English and Old Norse attest the complete 3rd class para­ digm. 13 In fact, my only reason for assuming the former existence of stems in *-e- in Germanic is that they provide a convenient way of accounting for word equations like Go. pahan = Lat. tacere 'be silent' or Go. jJulan 'endure' = Lith. tyleti. Forms like *pag-e- and *pul-e­ have left no direct trace in any Germanic language. 14 Indeed, it is perfectly possible that the productivity of the 3rd class may belong partly to a period later than the restructuring of the thematic middle as an active in 3sg. *-aijJ(i). In particular, note that if the phonological reflex of IE *-ej,eti is *-aijJ(i) in Germanic, then the membership of denominatives like OHG rot~n in the 3rd class could simply reflect the merger in the 3sg. of a thematic type in *-ej,eti (cf. Lith. Tudeja) with the renewed 3sg. middle in *-ai + *-p(i). This would make it unnecessary to assume the former existence of a middle paradigm for denominatives. 862 LANGUAGE, VOLUME 49, NUMBER 4 (1973)

It is a significant fact, as observed before, that 3rd class weak verbs in Ger­ manic commonly correspond semantically to middles in other IE languages (cf. Meid 1971:7). This is hardly surprising, for the 'internal' value of the IE middle (see especially Benveniste 1966: 168-75) is very close to the stative, generally intransitive, value of formations in *-e-: one is reminded of the devel­ opment of the Greek aorist passive in *-(th)e- from such pairings as phainomai '1 appear', ephdnen (aor.) 'I appeared'; trepomai 'I turn', etrdpen (aor.) '1 turned'. Among the Germanic 3rd class verbs which translate the middle or deponent presents of other languages, we may cite Go. sifan 'rejoice', Skt. hdr~ate, Lat. laetor; OHG *hog~ 'ponder, meditate', Gk. dianoeomai, Lat. meditor; Go. pulan 'suffer, endure', Skt. k~amyate, Lat. patior; OHG folg~ 'follow', Skt. sacate, Gk. hepomai, Lat. sequor etc. Further examples are not hard to find. Special interest, of course, attaches to those cases where a Germanic 3rd class weak verb can be related etymologically as well as semantically to middle formations elsewhere. We have already seen that Go. hahan, OHG hangbl, is such a verb; another may be the 'live' (OE libban, OS libbian etc.), which has a striking formal counterpart in the Tocharian 3rd class present lipefii,r 'is left, remains' < *lipotor. Yet another lexical item is 3sg. *wunaip(i) 'is content' (represented by OHG won~t, Go. unwunands 'unsatisfied' etc.), which it is tempting to compare with Skt. vdnate 'desires'; both verbs can be derived from a thematic middle prototype 3sg. *1Jon-6i, contrasting with the inherited active 3sg. *1Jon-neu-ti, Skt. van6ti 'overcomes', OE thematized winnep < *winwip 'struggles'. (On the semantics of the root van-, cf. Grassmann 1873, sub voce.) Important historical information is furnished by an archaic group of 3rd class weak verbs which occur beside preterit-presents formed from the same roots. This t·ype is well represented in Gothic: munaip 'thinks', man 'remembers'; -kunnaip (ga-kunnaip 'finds out', uf-kunnaip 'realizes' etc.), kann 'knows'; witaip 'watches', wait 'knows'; gaparbaip sik 'abstains', parf 'lacks, needs'. To these we may add *agaip 'takes fright', og 'is afraid' on the strength of the ad­ jective unagands 'unafraid', wrongly taken by Krause (1968:234) to represent a lost strong verb *agan (3sg. *agip). The former existence of several other such pairs can be concluded from the evidence of North and West Germanic forms: ON dugir 'helps, is useful' is the 3rd weak class counterpart to Go. impersonal daug 'is useful', and OHG lirnet 'learns', though with a nasal suffix, otherwise stands in the same relation to Go. lais 'knows, understands'. Besides confirming several of the Gothic pairs, OHG preserves the 3rd class weak verbs maget 'is strong', sculet 'owes', alongside the preterit-presents mag 'can', seal 'ought'. The co-occurrence of preterit-presents and 3rd class weak verbs cannot be motivated synchronically in Germanic, and must be regarded as continuing an old pattern. Wagner (passim) has explained weak verbs like *kunnaip(i), *dugaijJ(i) etc. as the product of a derivational process whereby stative stems in *-e- were formed to underlying perfects in late IE. This view is probably correct as far as it goes: within Germanic at least two pairs-Go. munaip, man and witaip, wait-corre­ spond term for term to e-statives and perfects in other branches of the family. The root *men- attests both formations in Gk. memona (= Lat. memini) vs. emdnen (post-Hom.); Lith. mini vs. mineti; OCS m'bnit'b vs. m'bneti; *1Jeid-, THE GERMANIC THIRD WEAK CLASS 863 beside its familiar perfect *1Joida 'I know' (Gk. (w)oida, Skt. veda etc.), shows the stative stem *1Jide- seen in Lat. video, Lith. pavyd-eti, OCS videti. But while the association of the perfect and e-stative may indeed date back to a very early period, it seems even more important in an IE context to take note of the rela­ tionship of the perlect to a third and older verbal category-one to which it may ultimately be etymologically identical. This is the IE middle. I have already alluded above (§2) to the common origin of the perfect and middle in IE. A mechanism whereby the two categories may originally have come to be differentiated is suggested by Kurylowicz (1964: 62-3); see further Watkins (1969: 112-3) and, for a somewhat different view, Meid (1971 :32-9). Although some of the details of the intervening stages lie outside our view and probably always will, the basic identity of the perfect and middle endings (in particular the pi. 3sg. in *-e and the middle 3sg. in *-0) is as firmly established a result as any IE morphology, and may be taken as a point of departure. The assumption of a common origin for both categories helps explain the fact that a significant number of verbs whose perfects are of IE date also preserve traces of an old middle inflection. Two striking examples are the roots *men- and *1Jeid­ just discussed. In Vedic, where both are well-attested, the most archaic non­ perfect forms are the aorists dmata and dvidat. Of these, the former is a hapax root-aorist middle, superseded in the later Vedic language by the sigmatic dma'Y(tsta; the latter is a thematic aorist, quite possibly the original representa­ tive of its type. The nature of the relationship between these two aorist forma­ tions has now been elucidated by Watkins (1969:88 ff. etpassim), who shows that both the athematic middle in -ta and the thematic active in -at presuppose an old middle 3sg. in *-~/6. Amata, like dmrta 'died', dgata 'went' etc., ultimately e e continues a 3sg. *mon-~/6 (*m'or- /6' *g'};om- /6)' which has been remade to *m'l'}-to (*mr-to, *g~1(t-to) via thenow familiar replacement *-0 ~ *-to. Thethematic aorist dvidat rests on a different renewal, the same as the one seen above in discussing the forms dduhat < dduha and dSayat < *d8aya. We may set up, with Watkins, a pre-Indic *dvida, or in IE terms a 3sg. *1Jid-~/6; this hypothetical form is in fact directly confirmed by the archaic dentalless 3sg. present vide 'is found'. For IE it is legitimate to assume perlects *(me)mdn-e, *1Joid-e and middles *mon-o, *1Jid-o (with the usual generalization of o-timbre in the middle ending). The assignment of the latter two forms to the aorist category in Sanskrit seems to have been a relatively late and, it would seem, incomplete development: in the case of *1Jid-o it was probably the aoristic nuance of the root itself, in that of *mon-o the encroachment of a specially characterized present *m'l'}-i//6- (Skt. manyate, cf. Gk. matnomai) which led to the specialization of fLvidal, drnata as aorists. In most of the remaining IE languages, the middles *mon-o, ·?Jid-o, unlike the comparatively stable perfects *(me)m6n-e, *1Joid-e, were eliminated or drastically restructured; and only in one branch of the family, Germanic, were they actually maintained as part of a coherent system in which the func­ tional opposition of perlect and middle is preserved intact. This opposition, I suggest, is still perceptible in Go. man 'remembers' vs. m'UnaijJ 'thinks' and Go. wait 'knows' vs. witaijJ 'watches', where the 'middle' terms munaijJ and witaijJ represent the normal Gmc. reflexes of presents *mon-oi and *1Jid-oi (= Vedic 864 LANGUAGE, VOLUME 49, NUMBER 4 (1973) vide), transformed by the process detailed in the ·preceding pages. The same contrast of state vs. what we may call 'internal activity' is typical of several other such pairs: 'know, understand' vs. 'learn', 'be able' vs. 'become able', 'lack'vs. 'abstain from' etc. Formally and functionally, 3rd class weak verbs of this type continue an IE category with remarkable fidelity. The Vedic present vide constitutes an important support for the assumption of a pre-Germanic 3sg. *witai (> Go. witaijJ etc.) Beside the double equation veda: vide = wait: witaip, I now propose to add a second, Vedic duduhe (2sg. active dud6hitha) : duhe = Go. daug : ON dugir. The etymological con­ nections of the Indo-Iranian root *dhugh- have long been disputed; most scholars have been hesitant on semantic grounds to accept an equation with Gmc. forms meaning 'be suitable, help' (see literature in Mayrhofer 1953). Yet the phono­ logical side of the comparison is unexceptionable, and the semantic objections have been persuasively ans,vered by Meid in his recent study of the Gmc. preterit. Meid's remarks (1971:24-5) on the IE root *dheugh- deserve to be quoted at some length: daug. Das Verbum ... stammt ab von idg. *dheugh-, dessen urspri.ingliche Bedeutung im weitesten Sinne etwa mit 'Ertrag produzieren, nutzbar machen' (Akt.), 'Ertrag gehen, Nutzen bringen' (Med.) angesetzt werden kann. Die Bedeutung impliziert den Begrift' der MENGE, FtlLLE, auch der QUALITAT. 1m ALTINDISCHEN, das hier fri.iheste idg. Vieh­ ziichterkultur wiederspiegelt, ist duh- ein Begriff der Milchwirtschaft und erst von da aus bildlich erweitert: duh- bedeutet ... a) aktivisch '(heraus)melken' (auch bildl.), b) 'Milch geben', iibertr. '(reichlich) stromen, spenden' u. dgl. Vgl. RV 3, 57,1: '(Wer sich meine Dichtung ausersehen hat, hat eine) Milchkuh (gefunden), die viel an Labung milcht': dhenurp ... yO, duduhe bh'i1ri dhaser. The peculiar sense of the root duh- is thus secondary, the result of its early in­ corporation into the expanding Indo-Iranian lexicon of terms connected with cattle breeding. The ON 3sg. dugir 'helps, avails' « *dugaijJ) continues, in the gulse of a 3rd class weak verb, the same isolated IE form *dhugh-Oi that survives almost unchanged in Skt. duke. In both Germanic and Indo-Iranian, the reten­ tion of the corresponding pf. *(dhe-)dhdugh-e (daug; *dud6ha, duduhe) is a characteristic archaism, entirely comparable to the survival of veda beside vide in Sanskrit, and wait beside witaijJ in Gothic. We can now appreciate the significance of the verbs just discussed for the 3rd weak class as a whole. Pairs like *yoide vs. *yid6(i), *m6ne vs. *mon6(i), *dhOughe VB. *dhugh6(i) are inherited, and ultimately reflect the common origin of the perfect and middle in IE. In a central group of languages including Ger­ manic, such pairs appear to have been augmented by a third series of forms built on a stem in *-e-: *Uide-, *mone-, *dhughe-. These forms were probably aorists; but while their original function in Germanic is impossible to determine exactly, we can be reasonably sure that they did NOT playa role in the formation of the presents *witaip(i), *munaijJ(i), *dugaijJ(i). It seems instead that, at some time in the pre-history of Germanic, the middle paradigms corresponding to 3sg. *Uid6(i) , *mon6(i), *dhugh6(i) were thematized owing to the identity of the thematic and athematic types in the 3sg.; from this point on, their history in the present system became indistinguishable from that of originally thematic middles like Go. intrans. hahan and (probably) ON heita. Thus these verbs sup- THE GERMANIC THIRD WEAK CLASS 865 ply an early set of examples in which a thematic middle present (3sg. *1Jid6i > *witai(jJi) etc.) came to stand beside a non-present form in *-e- elsewhere in an extended paradigm. Once established in the language, this pattern was free to spread to verbs like 'have', where the stem in *-e- is apparently old, but where middle inflection in the present can only be secondary. The OHG 3rd class weak verb lirntn, lernen (= OE liornian, also with 3rd class forms) belongs etymologically with the Gothic preterit-present lais 'knows, understands'; the two verbs constitute a pair that cannot be separated from those discussed above. Formally, however, CGmc. *liznaijJ(i) presents a prob­ lem: since IE deverbal *-e- was applied directly to roots rather than present stems, the seeming occurrence of a present in *-ne- is difficult to account for in traditional views of the 3rd class. In terms of the theory just put forth, nothing is more natural than to assume that *liznaijJ(i) continues an older *liznai, the form we would expect as the regular middle 3sg. to an IE nasal present (cf. Skt. grT)e 'is praised' < *g~r-n~2-6i beside active gr1jati < *g~r-ne~-ti). In the same fashion it is possible to derive Go. kunnaijJ (= OHG kunnet) from a pre-Germanic prototype *kunnai < *g1)-n~3-6i. This etymology has the advantage of accounting for the ·nn- of the historical forms without appeal to an ad-hoc phonetic rule, and at the same time permits us to explain the difficult forms Go'. kann, OHG kann etc. as a CGmc. back formation from *kunnaijJ(i), after the model of other such pairs. *LiznaijJ(i) and *kunnaijJ(i) are not the only 3rd class weak verbs with a nasal suffix. A small but significant number of others, such as OHG hlin~n (= OE hlinian) 'lean', OHG mornen 'grieve', and ON tolla 'stick to'-all intransitive-show fairly clearly that the type is old, and that a group of 3rd weak class nasal presents must be assumed for Common Germanic. Apart from their interest for the history of the 3rd class as a whole, these forms are important for a second reason: they may shed valuable light on the prehistory of another problematic category, the so-called 4th weak class of nasal inchoatives (type fra-lusnan 'become lost') in Gothic. The Gothic nasal inchoative class is characterized by a 'strong' present (3sg. fra-lusnip) beside a weak preterit formally indistinguishable from a preterit of the 2nd class (fra-lusnoda). While these verbs do not form a distinct conjugation in any other Germanic language, Old Norse has a large number of nasal presents like vakna 'wake up', losna 'become loose', and sortna 'become dark', which be­ long to the 2nd class and clearly continue the same original category as the Gothic type. On the strength of the Norse forms and the Gothic preterit, most scholars (e.g. Meid 1967:252-4; Krause, 246) have sought to derive the Germanic nasal inchoatives from the IE present type in *-ne~2-/-n'()- (whence *-na-/-n'd-, Gmc. *-no-/-na-) which underlies the Skt. 9th class (3sg. krirJiiti 'buys', 3pl. krirjanti). The Germanic and Sanskrit formations are clearly related in some way. Two facts, however, stand in the way of equating them directly: first, the Gothic present, unlike the IE type, is thematic; second, the Germanic forms are incho­ ative and intransitive, while the IE type is characteristically transitive and even causative (cf. Skt. rarnate 'rests', ram1Jati 'brings to rest'). Both the form and function of the Gothic 4th class can be explained on the assumption that the Gennanic nasal inchoatives were originally inflected as 866 LANGUAGE, VOLUME 49, NUMBER 4 (1973)

middles. We have just noted that Germanic preserves remnants of a group of nasal presents which belong to the third class. Several of these verbs, such as OE Jinian (OHG ginen) 'yawn' beside Jinan,15 or OHG mornen beside OE murnan, exhibit a tendency to appear with strong but synonymous by-forms. Within the context of the theory presented above, it is natural to interpret such doublets as the result of an original contrast, in at least some nasal presents, between active and middle forms; when Germanic subsequently lost the active vs. middle opposition, there resulted, in semantically favorable cases, synonymous or nearly synonymous pairs which could provide a model for other originally middle verbs to adopt active inflection.16 It is probable that the Gothic verbs in -nan, 3sg. -nip represent the result of just such a transfer. At the outset of the Ger­ manic development we may assume three present types: an athematic active in 3sg. *-nop(i) < IE *-ne~2-ti, e.g. *waknop(i) 'wakes' (tr.); a thematic active in 3sg. *-nijJ(i), probably thematized from the preceding type, e.g. *waknijJ(i) 'id.'; and a thematic middle in 3sg. *-nai < IE *-n~2-oi, e.g. ·waknai 'wakes up' (intr.) A-priori, we might expect that this state of affairs would be reflected in the attested stages of Germanic by three formations, two transitive and one in­ transitive. In fact, however, the existence of isofunctional pairs like OE )inian vs. Jinan, OHG mornen vs. OE murnan, together with the expansion of the iterative-causativetypein *-ja- < IE *-eie/o- (*wakijJ(i) > Go. wakjip 'wakes', tr.) led to a breakdown in the inherited system of oppositions. The nasal suffix, replaced in its transitive functions by the formally unrelated causative suffix, was re-interpreted as a mark of intransitivity, thereby rendering the 3rd class (originally middle) inflection of intransitive-inchoatives entirely redundant. The 3rd class presents were thus free to be replaced by the inherited active types in 3sg. *-nop(i) and 3sg. *-nip(i), Old Norse in general preferring the former al­ ternative and Gothic the latter. In the more or less confused situation of the attested languages, the nasal presents continue three distinct types in form---­ but only one, the middle, in function. 17 4. In the preceding pages I have tried to show that an archaic core of 3rd class weak verbs can most easily be derived, at least in the present system, from a Common Germanic transformation of the IE middle. Before concluding this study, I will briefly survey, language by language, the major formal innovations which have occurred between the Common Germanic period and the time of our written records. The point of departure, it will be recalled, is the type presented in §2, with endings Isg. -0, 2sg. -ais(i), 3sg. -aijJ(i), Ipl. -ame(s), 2pl. -aijJ(e), 3pl. -anjJ(i).

16 The long i here (and in the rhyming strong verbs cinan 'gape', dwinan 'w88te away', scinan 'shine') is surely secondary; perhaps it arose by back formation from the earlier forms of the preterits glin, can, dwan, scan, either before or after these had acquired their nasal from the present. 16 Pairs of this kind, of course, are not confined to the nasal presents. We may assume a transfer of the same kind in Go. bauan 'live, dwell', strong pres. 3sg. bauijJ, but 3rd class weak prete bauaida. The weak pres. *bauaip is still normal in East Norse and has left traces elsewhere in Germanic (see Flasdieck, 89-93, for an inventory of forms). 17 On the semantic overlap of the IE middle with the Gothic 4th class, see now Meid (1971: 7). THE GERMANIC THIRD WEAK CLASS 867

GOTHIC: The Gothic 3rd class paradigm preserves the Common Germanic situation essentially unchanged. The only analogical form in the present is the 2sg. impv. habai (for phonologically regular *haba) , probably created to 3sg. habaip on the model of the 2nd class impv. salbo beside 3sg. salbojJ. The optative 2sg. habais, 3sg. habai etc. is indistinguishable from that of any other thematic present, and continues the Common Germanic type intact; note that, since I have assumed the exclusion of *-e- from the present system, there is no need to envisage an athematic optative *habe-i-s, *habe-i etc. An archaic feature seems to be preserved in the anomalous prete 3sg. uf-kunjJa to (weak) uf-kunnan 'recognize'; cf. OS habda, sagda etc., likewise formed without a connecting vowel. The regular type habaida, although perhaps already Common Germanic, rests on an analogy with the other weak classes: habaip: habaida = salbojJ : salboda = *nasijJ (replaced by nasiijJ) : nasida. OLD NORSE: The treatment of the 3rd weak class in North Germanic is quite conservative: in verbs of the normal type only the 1sg. in -i is not an inherited form; and this, as already seen, reflects a very general Norse repatterning. The three verbs hafa, segia, and jJegia are conjugated irregularly. Hafa shows umlaut in the singular, but not in the plural: lsg. hef, heft, 2-3sg. hefr, hefir, 1pl. hQfum, 2pl. hafi'0, 3pl. hafa. In the longer sg. forms heji, hefir, it is simplest to see the result of a contamination of the short forms hef, hefr with the regular 3rd class forms *haft, *hafir. (These latter have also played a role in the formation of haver < *hajr; cf. Flasdieck, 117-22.) The remaining present forms of hafa, in particular hef, hefr and the entire plural, are indistinguishable from those of a normal strong verb of the 6th class. From a Norse point of view, it is simplest to assume the original presence of two paradigms, one originally middle (*habo, *habais, *habaijJ etc.), the other originally active (*habo, *habis, *habij; etc.); the relation between the two would be entirely comparable to that between pre-Norse *buaip and *bUip 'lives' or pre-English *ginaip and *ginip 'yawns'. Note that there is no unambiguous Norse evidence for a stem *habia-. In North Germanic, stems in *-ja- play a role in the 3rd weak class only insofar as they are needed to account for the forms of segia and its rhyming companion pegJa, which have the same forms as hafa in the singular but resemble verbs of the 1st weak class in the plural (lpl. segium, 2pl. segi'O, 3pl. segia). The absence of gemina­ tion in the plural and the long forms segi, segir in the singular point to a mixture of two inherited paradigms-*sago, *sagais, *sagaijJ etc. (3rd class), and *sagio *sagis, *sagip etc. (1st class). The latter forms will be discussed below in con­ nection with the situation in Old English and Old Saxon. OLD HIGH GERMAN has drastically simplified the inherited 3rd class paradigm by generalizing the stem *habai- (> OHG habe-) and utilizing it to create a new athematic paradigm modeled on that of salb6n (cf. 1sg. habhn, opt. habee etc.) The isolated 1sg. forms habu, sagu (Tatian) are the only relics of the old system in the present; in the preterit, hapta (cf. OS habda) is archaic beside regularized hab~ta and secondary hebita. The difficult 2sg. and 3sg. forms hebis hebit, segis segit, libis libit are probably to be taken at face value and compared with ON hefr, segr. This would imply that OBG, like Old Norse, inherited al­ ternate paradigms *habo, *habis, *habijJ etc. and *sagio, *sagis, *sagij; etc. 868 LANGUAGE, VOLUME 49, NU~fBER 4 (1973)

(*libjo, *libis, *libijJ etc.) beside the normal 3rd class presents *habo, *habais, *habaijJ etc. Ad hoc as it may appear, this interpretation of the OHG facts seems far preferable to assuming that hebis, hebit etc. somehow reflect an older paradigm *habjo, *habais, *habaijJ etc. (as maintained, e.g., by Flasdieck, 157-8) in which the 1sg. *habjo (> *hebbu) gave rise to analogical *habis, *habijJ (> hebis, hebit), then disappeared ,vithout a trace. In any case, note that neither habhl" saghl" Leben nor any other OHG verb provides direct evidence for an original alternation between stems in *-ja- and *-ai- within a single paradigm. Pace Flasdieck, a form without a-umlaut like lirnen (for phonologically regular lernen) can as easily be attributed to a lost 3sg. liznijJ (cf. Go. fra-lusnijJ) as to a questionable lsg. *lizn(i)jo. OLD ENGLISH AND OLD SAXON: Very little remains of the 3rd class as an autonomous category in Northern West Germanic. We have already seen how Old English has created a new type in 1sg. *-aijo (lsg. lif3e < *libaijo beside libbe), which itself loses ground to the expanding 2nd class. The 2nd class has expanded enormously at the expense of the 3rd in Old Saxon also; as in Old English, only the three verbs 'have', 'say', 'live' retain any features of mor­ phological interest. The characteristic innovation of both languages is the replace­ ment of simple thematic forms like 1sg. *sago, pI. *saganp etc. by corresponding forms in *-ja-: OE secge secgap; OS seggiu, seggiad < *sagjo, *sagjanp. In my opinion, these ja-forms belong properly not to the 3rd class, but to the alternate paradigms *sagjo, -is, -ip etc., and *libjo, -is, -ijJ etc., attested in­ directly in Old Norse (*sagjo only) and inOIIG. (The pattern of these two verbs seems to have been extended to produce *habjo, -is, -ijJ etc. in Northern West Germanic, as against pre-Norse *habo, -is, -ip etc.) 1:'he co-occurrence of 1st and 3rd class presents in the same verb is not confined to the handful of verbs under discussion, but recurs in such pairs as Go. hugjan 'ponder' vs. OHG *hogen (pret. hogeta; cf. hycgan and hoJian in Old English), ON fylgja 'follow' vs. OHG jolgen, Go. hatjan 'hate' (cf. OE hettend 'enemy') vs. hatan, OHG hazzen. Here the 1st class present probably reflects the IE type in *_j6/0- which can be seen in Ok. mainornai, Skt. pu~yati 'prospers' etc. Within the present framework, the simplest way to account for the twofold inflection of *hug-, *fulg-, and *hat­ is to assume that they originally opposed a non-present stem in *-e- to an in­ herited present with the suffix *-ja-; in keeping with the productive Germanic pattern, the 'aorists' *huge, *fulge, *hate would then have given rise to new and regular 3rd class presents *hugaijJ(i), *fulgaijJ(i), *hataijJ(i). Some similar set of circumstances may well be responsible for the creation of the doublets *libjo, *libo, and later, *habjo, *habo; in the case of *sagjo, *sago, the forms in *-ja- can most plausibly be taken from an iterative *sok'/:eio. 18 In each case the link con­ necting the two presents must have been the non-present stem in *-e-. The subsequent fate of the three verbs 'have', 'say', and 'live' has been dictated in large part by the fact that, in the inherited 3rd class paradigm, the two stems in *-ai- and *-a- (West Germanic *-e- and *-a-) are, from a descriptive point of view, suppletive-i.e., they cannot be predicted from each other on the basis

18 This suggestion is due to Patrick Hollifield. THE GERMANIC THIRD WEAK CLASS 869 of any other morphophonemic alternation in Germanic. Given the gradual re­ duction of the 3rd class to a mere three lexical items, and the existence of com­ peting 1st class paradigms for each of these, it is not hard to see why forms in *-ja- have tended to become more common. In the 1sg. (and, mutatis mutandis, the 3pl., infinitive etc.) neither *sago nor *sagjo is 'regular' with respect to 3sg. *sagaijJ. *Sagjo, however, has at least the advantage of being normal in a produc­ tive class of weak verbs; it has spread at the expense of *sago, replacing this form in Old Saxon and the 'Saxon' dialects of Old English (but cf. Northumbrian sago). By the same token, the 2sg. and 3sg. may be replaced without reference to the form of the 1sg.: Old Saxon (Heliand C) has habis habit, sagis sagit-forms which cannot be separated from OHG hebis hebit, segis segit, and where the absence of umlaut is clearly the result of contamination with the proper 3rd class forms habes -as, habed -ad (Heliand M). In Old English the 2sg. and 3sg. forms hafas(t) hafajJ, sagas(t) sagajJ, leofas(t) leo/ajJ show a different substitution, reflecting the now familiar tendency of the 2nd class to encroach upon the third; one may compare further OS libos, libot for *libes -as, *libed -ad. These 2nd class forms, like those in *-J'a-, are comparatively late intruders in the 3rd class paradigm. The manner in ,vhich 1st and 2nd class forms have penetrated into the normal inflection of the verbs 'have', 'say', and 'live' is instructive: even when inherited 3rd class forms are eliminated, the new forms which replace them preserve intact the archaic opposition of two synchronically unrelated stems, one characteristic of the 2sg. and 3sg., the other of the remaining members of the paradigm.

REFERENCES BENNETr, WILLIAM H. 1962. The parent suffix in Gennanic weak verbs of class III. Lg. 38.135-41. BE:NVENISTE, l'TIMILE. 1966. Actif et moyen dans Ie verbe. Problemes de linguistique generale, 168-75. Paris. CHANTRAINE, PIERRE. 1925. Les verbes grecs en tho. Melanges linguistiques ofierts a M. Joseph Vendryes par ses amis et ses eleves, 93-108. Paris. COWGILL, WARREN. 1959. The inflection of the Gennanic a-presents. Lg. 35.1-15. --. 1968. The first person singular medio-passive in Indo-Iranian. Pratidanam: Indian, Iranian, and Indo-European studies presented to F. B. J. Kuiper, ed. by J. C. Heestennan et al., 24-31. The Hague. FLASDIECK, HERMANN. 1935. Untersuchungen fiber die gennanischen schwachen Verben III. Klasse. Anglia 59.1-192. GRASSMAN, HERMANN. 1873. Worlerbuch zum Rigveda. Leipzig. JASANOFF, JAY. 1968. Studies in the historical morphology of the Hittite verb. Harvard dissertation. KRAHE, HANS. 1961. Altgennanische Kleinigkeiten. IF 66.35-43. KRAUSE, WOLFGANG. 1968. Hal1dbuch des Gotischen. Munich. --, and HERBERT JANKUHN. 1966. Die Runeninschriften im alteren Futhark. Got- tingen. --, and WERNER THOMAS. 1960. Tocharisches Elementarbuch I. Heidelberg. KURYf:,OWICZ, JERZY. 1956. L'apophonie en indo-europeen. Wroclaw. --. 1960. Esquisses linguistiques. Wroclaw. --. 1964. The inflectional categories of Indo-European. Heidelberg. MAYRHOFER, MANFRED. 1953. Kurzgefasstes etymologisches Worterbuch des Altindischen. Heidelberg. 870 LANGUAGE, VOLUME 49, NUMBER 4 (1973)

MElD, WOLFGANG. 1967. Gennanische Sprachwissenschaft, III: Wortbildungslehre. Berlin. --. 1971. Das germanische Prateritum. Innsbruck. NOREEN, ADOLF. 1923. Altnordische Grammatik, II. Halle. POLOME, EDGAR. 1967. On the origin of the Germanic class III of weak verbs. Beitrage zur Indogermanistik und Keltologie: Julius Pokorny zum 80. Geburtstag, ed. by Wolf­ gang Meid, 83-92. Innsbruck. PREVOT, ANDRE. 1935. L'aoriste grec en -then. Paris. SCHMID, WOLFGANG P. 1963. Studien zum baltischen und slavischen Verbum. Wiesbaden. WAGNER, HEINRICH. 1950. Zur Herkunft der e-Verba in den indo-germanischen Sprachen. Zurich. WATKINS, CALVERT. 1969. Indogermanische Grammatik, 111.1: Geschichte der indo­ igennanischen Verbalflexion. Heidelberg. --. 1972. The Indo-European denominative statives in *-e-. To appear in TPS. [Received 3 November 1972.]